How Akamai and NVIDIA Are Embedding Zero Trust Into AI Factories

As autonomous AI agents begin to handle sensitive enterprise data and make consequential decisions, the infrastructure that powers them must be secure at every layer. Traditional perimeter-based security—a firewall here, a software agent there—falls apart when agents move across distributed inference pipelines, share context memory, and access files dynamically. The industry needs a model where every action is verified, and trust is never assumed. That model is Zero Trust, and two infrastructure heavyweights—Akamai and NVIDIA—are embedding it directly into AI hardware.

On June 2, 2026, the companies announced a deepened collaboration that places Akamai’s Guardicore Segmentation software into NVIDIA’s Vera BlueField-4 STX storage architecture, running on the DOCA platform . The goal: protect the data, context memory, and autonomous agents that form the core of modern AI factories . This article breaks down what was announced, how the integration works, the prior collaboration that set the stage, and when to expect these solutions.

The June 2026 Announcement: Zero Trust Inside the AI Factory

The new partnership centers on integrating Akamai Guardicore Segmentation with the NVIDIA Vera BlueField-4 STX, a Data Processing Unit (DPU) designed for the storage layer of the NVIDIA Vera Rubin platform . By running Guardicore’s segmentation engine directly on BlueField-4 silicon via the DOCA software stack, the solution enforces a Zero Trust cybersecurity policy at the hardware level—without the need for software agents on every workload .

The announcement frames this as a natural extension of the agentic AI era: as autonomous agents become the “workers” inside AI factories, the data paths they use, the memory they share, and the files they access all become potential attack surfaces . The integration is designed to continuously validate each of those interactions, ensuring that only authorized workloads see sensitive data and that compromised agents can be isolated immediately .

How the Integration Works

The technical integration combines three layers of protection, all governed by Guardicore’s policy engine and enforced in BlueField-4 hardware .

Silicon-Enforced Segmentation

Instead of installing agents on every virtual machine or container, segmentation policies are enforced inline on the DPU itself. This “agentless” model inspects traffic and file access at line speed—up to 800 Gb/s—without consuming GPU or CPU resources needed for training and inference .

Three Key DOCA Security Capabilities

• DOCA Vault acts as a hardware root of trust for file access. It cryptographically ensures that only authorized AI workloads can open the right files with the correct permissions .
• DOCA Argus provides visibility into agent and workload behavior. It monitors what autonomous agents are doing, how they’re interacting with data, and whether their behavior deviates from policy .
• DOCA Flow handles network-level isolation. In multi-tenant AI environments, it creates micro-segmented zones to prevent lateral movement between workloads, limiting the blast radius if one is compromised .

Real-Time Protection for Agentic Workloads

An agentic AI workload involves a chain of reasoning, memory retrieval, tool use, and inter-agent communication that spans multiple infrastructure components. The integrated platform inspects and governs each interaction—between agents, between agents and data, and between agents and context memory—inline . When Guardicore’s visibility engine identifies a threat pattern, the hardware enforces a policy decision in real time, without requiring a separate enforcement point outside the data path.

NVIDIA reports that this approach delivers runtime threat detection up to 1,000 times faster than existing agentless runtime solutions . This speed difference matters in AI contexts where an adversary exploiting a compromised agent can exfiltrate context memory or inject malicious instructions in microseconds.

Roots of the Partnership: The February 2026 OT/ICS Solution

The June AI factory announcement didn’t come out of nowhere. On February 23, 2026, Akamai and NVIDIA revealed their first joint security offering: an agentless Zero Trust segmentation solution for operational technology (OT) and industrial control systems (ICS) .

That earlier collaboration paired Akamai Guardicore Segmentation software with NVIDIA BlueField DPUs—the prior-generation data processing units—to protect “un-agentable” equipment in power plants, water facilities, and manufacturing floors . The problem in OT environments is acute: legacy industrial machinery often cannot run traditional security software because installing an agent would disrupt operations or simply isn’t supported. The jointly developed solution offloads all security processing to the BlueField DPU, creating a hardware-isolated security layer that operates independently of the protected devices .

The February announcement positioned this as a way to align critical infrastructure with evolving cybersecurity regulations while maintaining performance and uptime . It also marked the beginning of a broader NVIDIA effort to embed Zero Trust across multiple partner ecosystems .

Availability: When to Expect These Solutions

The timeline for each collaboration reflects the differing maturity of the underlying hardware.

• February 2026 OT/ICS solution: Expected to be globally available in Q2 2026 .
• June 2026 AI factory solution: Expected in the second half of 2026 (H2 2026), aligning with the broader Vera Rubin platform and BlueField-4 silicon ramp .

The staggered rollout shows a deliberate strategy: prove the agentless Zero Trust model on existing BlueField DPUs in industrial environments, then scale it to the more complex, data-intensive world of AI agent workloads as the next-generation silicon arrives.

Why This Matters for Enterprise AI

For organizations deploying autonomous AI agents in production, the security model must move at the speed of the agents themselves. Traditional agent-based tools can’t keep up—and in some architectures, they can’t even be installed. Hardware-enforced Zero Trust, powered by a unified policy engine and enforced at the infrastructure layer, offers a path toward security that doesn’t compromise agent performance or coverage.

By embedding segmentation directly into the storage and networking fabric of the AI factory, Akamai and NVIDIA are building a model where security is a property of the infrastructure, not an afterthought bolted on at the edge. The real test will come when agentic AI deployments move from pilot programs to enterprise-scale production in late 2026 and beyond.